The AI recommends them. It just never brings them up.

I ran the engagement through the Resource-Based View and Dynamic Capabilities. That sounds like academic scaffolding and it changed the recommendations materially.

Under RBV, the question is not which tactics to run but which assets compound and resist copying. Domain authority, topical depth, community trust and a longitudinal user-data asset all do. That reframes SEO and AEO spend as building something the client keeps, rather than renting attention — which is the argument that justifies deferring paid acquisition entirely while Authority Score sits at 17.

Dynamic Capabilities supplied the timing argument. The move from search to answer engines is exactly the kind of environmental discontinuity where a small, fast operator can take ground before scale players consolidate. Endo45's disadvantage in size is, briefly, an advantage in speed.

Site Health scored 81% and AI Search Health 78%, which reads fine until you go page by page. Of 33 pages crawled, exactly one was fully clean. Twenty-seven carried at least one issue, four were redirects and one was broken — 16 errors and 54 warnings in total.

The specifics mattered more than the score: 12 duplicate title tags, 20 pages with no meta description, 17 pages with multiple H1s, and 660 links carrying no anchor text. No llms.txt at the root and no schema anywhere, which means nothing on the site told an answer engine what kind of content it was looking at or who had reviewed it.

One finding cut against the gloom. Across a seven-day Australian position-tracking campaign on 21 keywords, visibility rose 9.29 percentage points to 14.66% — driven almost entirely by a single fatigue-and-caffeine cluster and largely one page. Small, caveated, and the first evidence that this content ranks when it targets the right kind of query.